embedding-condensation

Measure layer-wise token embedding cosine similarity, assessing the severity of embedding condensation. Concept from [ICML 2026] Dispersion loss counteracts embedding condensation and improves generalization in small language models.

embedding-condensation has been downloaded 4,006 times in total on PyPI, including 3,075 in the last 30 days. The latest version is 2.0, released May 17, 2026.

Version2.0
Downloads
4.01k
LicenseNon-Commercial License — see ../LICENSE.md in the LM-Dispersion repository.
AuthorChen Liu
UpdatedMay 17, 2026

Downloads

Weekly, last 90d.
Includes CI traffic.

VersionsTotal2.*1.*
Range
View
Granularity
Group by
CI traffic
Stack: OffCI: Included3 / 5 series
Selected total7.26k
4.0kAll-time
3.1kLast 30 days
19Last 24 h
<0.01/sPer second

Version distribution

Share of downloads by released version. Computed over the last quarter.

  • 01

    2.0

    1.4k downloads
    35.2%
  • 02

    1.1

    1.1k downloads
    26.3%
  • 03

    1.0.0

    789 downloads
    19.7%
  • 04

    0.1.0

    753 downloads
    18.8%

Guess the next day

Thirteen recent days of embedding-condensation downloads. Drag the green handle on the right to guess where day fourteen lands.

TRUTH1958WEDTHUFRISATSUNMONTUEWEDTHUFRISATSUNMON